Huzaifa Ahmad

(a) pipeline to catch faulty predictions by models (b) object detection via Lidar point clouds (d) Test pipeline architecture and trained ML models (2D & 3D) (e) Configuration, Intrinsic and extrinsic calibration of cameras and Lidars (f) Testing and validation of the system onsite. (g) Scanning, Registering, and generation of complete end-to-end point cloud and blended map of the facility

(a) R&D for classical speech enhancement/ Noise suppression techniques (b) implementation of some of the vital research papers following the classical approach in noise reduction and speech enhancement. (c) experiment with domain-based Noise reduction/speech enhancement algorithms

(a) Developing project pipelines using state of the art ML algorithms for the tasks like Image classification, Object detection and Anomaly Detection using ResNet series, YoloV7, PatchCore and PaDiM (b) Developing python based small software tools for data preprocessing (c) Data mining and data labeling using multiple tools. Labeling include reference point labeling, error labeling, bounding box labeling and segmentation mask labeling

Brain tumor segmentation using SegNet and GANS, a smart approach to synthesize MRIs and labels similar to BraTS dataset (GAN), combine with the original dataset to decrease standardized data scarcity, hence, increased the Jaccard similarity index in segmentation up to 4%.

Visual saliency prediction, a project that compared saliency maps for SalGAN, Sam-ResNet, and ML-Net with DeepGaze II using matrices like KL Divergence and AUC, etc. Sam-ResNet outperformed others with a KL divergence score of 6.42.

Image Denoising, aimed at removing different complex noises from images using simple kernels and filters like Gaussian and bilateral and complex CNN-based algorithms as ResNet and FFDNet